No:  IEEE-754r and the C99 standard both say clearly that atanh(1) should be infinity 
and atanh(-1) should be -infinity, and furthermore that the 'divide-by-zero' exception 
should be raised rather than the 'invalid' exception.  It's a singularity, just like 
log(0).  (This makes even more sense viewed from the perspective of complex 
arithmetic, where atanh is defined at all points in the complex plane except -1 and 1, 
where it has log-type singularities.)

The general idea is that it's meaningful to set atanh(1) = infinity because that's 
what the limit of atanh(x) is as x approaches 1 from below;  similarly for atanh(-1) 
and log(0).